What: Detecting deforestation in the tropical rainforest
Where: Ethiopia, Indonesia, Vietnam, Ivory Coast
When: 2020 - 2021

Deforestation is a significant contributing factor to climate change. Using satellite imagery and machine learning, we are able to visualise where forest areas are threatened and disappearing. To achieve this, the Organisation Everitas developed a new algorithm that is used to detect and track both large and small-scale deforestation.

Concluding the project, we were able to prove that current global models overestimate tree cover loss events by between 65% to 400% by inaccurately flagging losses on plantations. We have proven the ability of the new Enveritas model to discern between true forest and plantation and can thus correct for this error. Using high-resolution imagery and detailed analysis, the model also made advances in detecting granular changes to the forest canopies and was able to detect small scale encroachment – which plays a large role in many developing countries and usually goes unnoticed.

The project showed that the high variation of deforestation accuracy and tendency to overestimate the deforested area in global models like Global Forest Watch (GFW) makes it unsuitable for tracking deforestation across time and region. Instead, the data shows a need to develop more region-specific models coupled with robust ground-truthing to accurately assess the extent of deforestation as well as to find solutions combatting its occurrence and effects.

Our partner for this project was the international NGO Enveritas which is based New York. As an organisation, they focus on comprehensive sustainability assessments of value chains in the coffee and cocoa sectors. Thus, after our previous engagement with them, where we surveyed the sustainability of Ethiopian coffee production, we are happy to conclude another project by working with their team, comprised of long-time experts in development cooperation, innovative thinkers as well as tech enthusiasts.

Download the whitepaper